I forgot to lock the scriptResult in the overview handler, and
apparently a cmd/go test is incidentally fetching the overview page at
some point during test execution, triggering the race.
This race was caught almost immediately by the new
linux-amd64-longtest-race builder (see
https://build.golang.org/log/
85ab78169a6382a73b1a26c89e64138b387da217).
Updates #27494.
Change-Id: I06ee8d54dba400800284401428ba4a59809983b2
Reviewed-on: https://go-review.googlesource.com/c/go/+/449517
Run-TryBot: Bryan Mills <bcmills@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Bryan Mills <bcmills@google.com>
Reviewed-by: Michael Matloob <matloob@golang.org>
"path/filepath"
"strings"
"testing"
+ "time"
)
var (
}
srv := httptest.NewServer(s)
+ // To check for data races in the handler, run the root handler to produce an
+ // overview of the script status at an arbitrary point during the test.
+ // (We ignore the output because the expected failure mode is a friendly stack
+ // dump from the race detector.)
+ t.Run("overview", func(t *testing.T) {
+ t.Parallel()
+
+ time.Sleep(1 * time.Millisecond) // Give the other handlers time to race.
+
+ resp, err := http.Get(srv.URL)
+ if err == nil {
+ io.Copy(io.Discard, resp.Body)
+ resp.Body.Close()
+ } else {
+ t.Error(err)
+ }
+ })
+
t.Cleanup(func() {
// The subtests spawned by WalkDir run in parallel. When they complete, this
// Cleanup callback will run. At that point we fetch the root URL (which
status := ""
if ri, ok := s.scriptCache.Load(rel); ok {
r := ri.(*scriptResult)
+ r.mu.RLock()
+ defer r.mu.RUnlock()
+
if !r.hashTime.IsZero() {
hashTime = r.hashTime.Format(time.RFC3339)
}